Spaces:
Running
Running
| title: Prime Rush — an agent environment built from a shipped battle royale | |
| emoji: 🎯 | |
| colorFrom: gray | |
| colorTo: yellow | |
| sdk: static | |
| app_file: index.html | |
| pinned: false | |
| license: apache-2.0 | |
| short_description: A live battle royale as an agent environment | |
| tags: | |
| - agents | |
| - mcp | |
| - benchmark | |
| - games | |
| - tool-use | |
| # Prime Rush Environment | |
| An agent environment built from Prime Rush, a battle royale in live service. | |
| An LLM agent takes over a player part-way through a live match and is scored on | |
| what it does next. The write-up, the stack, the costs and the scorecard are on | |
| the Space page; `index.html` is that page. | |
| Also here: | |
| - `indus_mcp/` — the MCP adapter over a live match | |
| - `tasks/` — Archipelago task pack for zone survival | |
| - `grading_eval/` — the reference grader | |
| - `tests/` — end-to-end tests against a stand-in server | |
| - `results/summary.json` — the numbers behind the tables | |
| Raw per-episode data, including every tool call and response, is in the | |
| [results dataset](https://huggingface.co/datasets/datagame/prime-rush-results). | |
| The game server is ours and is not public. Get in touch for access. | |
| Apache-2.0. | |